Writing study: Disentangling Latent Risk Pathways via Bayesian Hypergraph Inference

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Shengxian Ding, Haonan Gao, Pangpang Liu, Xinyuan Tian, Yize Zhao · Official ICML 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Architecture & representation (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

22 physical pages

Before references / appendix

≈ 9.1 page-equivalents

Main figures / tables

4 / 6

Attached appendix figures / tables

7 / 4

Attached appendix span

≈ 9.91 page-equivalents

Before references / appendix: ≈ 41.36%References / intervening material: ≈ 13.59%Attached appendix: ≈ 45.05%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1. Introduction	1	≈ 1.22 pp.	5.55%
2. Related Work	2	≈ 0.7 pp.	3.18%
3. Bayesian Hypergraph Pathway Model	3	≈ 1.43 pp.	6.5%
4. Scalable Structured Variational Inference	4	≈ 1.14 pp.	5.18%
5. Experiments	5	≈ 3.57 pp.	16.23%
6. Conclusion and Discussion	9	≈ 0.46 pp.	2.09%
Acknowledgements	9	≈ 0.16 pp.	0.73%
Impact Statement	10	≈ 0.05 pp.	0.23%
References	10	≈ 2.99 pp.	13.59%
A. Variational Inference Details	13	≈ 4.41 pp.	20.05%
B. Additional Simulation Results	17	≈ 2.34 pp.	10.64%
C. Additional Results for UKB Application	19	≈ 2.25 pp.	10.23%
D. Computational Cost	22	≈ 0.91 pp.	4.14%

All detected figure and table locations

Main text

• Figure 1 · p. 3
• Table 1 · p. 6
• Table 2 · p. 6
• Table 3 · p. 6
• Figure 2 · p. 7
• Table 4 · p. 7
• Figure 3 · p. 8
• Table 5 · p. 8
• Table 6 · p. 8
• Figure 4 · p. 9

Attached appendix

• Figure 5 · p. 14
• Figure 6 · p. 17
• Figure 7 · p. 18
• Table 7 · p. 18
• Figure 8 · p. 19
• Figure 9 · p. 19
• Figure 10 · p. 20
• Table 8 · p. 20
• Table 9 · p. 20
• Figure 11 · p. 21
• Table 10 · p. 22

Argument and organization

1. 1 Introduction; 2 Related Work

   Separates rare-outcome prediction from overlapping risk-structure discovery.

2. 3 Bayesian Hypergraph Model; 4 Structured Inference

   Builds the observation model, nested indicators, repulsion prior and dependency-preserving approximation.

3. 5 Experiments; 6 Conclusion and Discussion

   Uses simulation for structural truth and UK Biobank for prediction and associative interpretation.

Problem, claim and contribution

The introduction states that prediction alone is insufficient, motivating separate endpoints for latent recovery, uncertainty and predictive accuracy.

Introduction, p.1; experiment rationale, p.5 · Main paper

Argument and section order

The logical zero constraints are introduced in the generative hierarchy and explicitly preserved in the variational family, tying inference design to the model.

Sections 3.5-4.2, pp.4-5 · Main paper

Figures, captions and visual explanation

The workflow maps covariates through uncertain hyperedges to outcomes; a dense global network is paired with two enlarged pathway examples, with threshold and edge-width meaning in the caption.

Figure 1, p.3; Figure 4, p.9 · Main paper

Tables, comparisons and evidence

Repulsion is tested for structural stability rather than claimed necessary for prediction, and predictive ECE is separated from simulated structural-PIP calibration.

Table 3, p.6; Table 4, p.7; Appendix B.4, p.18 · Main paper

Appendix and reproducibility

The supplement reports CAVI versus MCMC calibration checks and separates the considerable discovery-time cost from fast prediction with hardware and memory figures.

Appendix B.4, p.18; Appendix D, p.22 · Main paper

Clarity and writing conventions

The conclusion explicitly calls the learned structure associative, an important limit on earlier pathway, etiology and mechanism language.

Conclusion and Discussion, p.9 · Main paper

What to reuse—and what not to infer

Transferable lesson: Match each interpretability claim to its own structural diagnostic, and distinguish predictive calibration from uncertainty over latent structure.

Boundary: Repulsion-based stability is not by itself a uniqueness theorem; one-cohort associative pathways and low inference latency do not establish causal mechanisms or clinical readiness.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-07

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan of the main argument through physical p.9, with focused physical-page text inspection on 1, 3, 4, 5, 6, 7, 8, 9. Framing, section progression, mechanism/proof explanation, displays, comparison conditions and conclusions were inspected. This is not exhaustive technical reading, proof verification or reproduction.

Appendix / supplement

Appendix organization and main-text pointers were checked; focused text inspection of physical pages 13, 18, 22. Remaining appendix proofs and details were not exhaustively read. No separately linked supplement inspected.

Visually inspected PDF pages

3, 9

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Published proceedings PDF, PMLR volume 306, retrieved 2026-10-07 · 22 pages